Method for rapidly predicting moisture content of sludge cake

Abstract

The invention relates to a method for rapidly predicting the moisture content of a sludge cake, and belongs to the field of sludge treatment. The method comprises the following steps: S1, measuring the sludge specific resistance of a plurality of sludge samples, conditioning the tyrosine content and tryptophan content in a dehydration filtrate and the water content of a filter-pressed mud cake, and constructing an original data set; s2, after an original data set is preprocessed, an input component and an output component are imported into a BR-BP neural network for training; s3, training the BR-BP neural network, adjusting parameters by adopting a Bayesian regularization optimization algorithm, selecting an optimal parameter in combination with the evaluation index, and then ending training to obtain a pre-trained BR-BP neural network; and S4, transmitting the input component of the to-be-predicted sludge to the pre-trained BR-BP neural network, and outputting a predicted value of the water content of the sludge cake. The problem that the actual requirements of engineering for precision and stability are difficult to meet due to the fact that prediction deviation is likely to occur under the condition of no manual parameter adjustment is solved.

Description

Technical Field

[0001] This application relates to the field of sludge treatment technology, and in particular to a method for rapid prediction of the moisture content of sludge cake. Background Technology

[0002] With the acceleration of urbanization in my country and the widespread construction of sewage treatment facilities, the output of municipal sludge has been increasing year by year. It is estimated that the annual output of wet sludge in China will exceed 80 million tons by 2025. Sludge dewatering is a key step in achieving sludge reduction, stabilization, harmlessness, and resource utilization. The moisture content of the sludge cake directly affects the choice of subsequent incineration, landfill, or resource utilization and the associated costs.

[0003] However, sludge dewatering performance is affected by multiple factors such as organic matter composition, floc structure, surface charge, and bound water content. Traditional evaluation indicators, such as sludge specific resistance (SRF) and capillary absorption time (CST), have limitations such as long measurement cycles, complex operation, and difficulty in online monitoring. Recent studies have found that aromatic proteins (tyrosine and tryptophan) in sludge extracellular polymeric substances (EPS) are key molecular markers affecting dewatering performance, and their content is significantly negatively correlated with sludge dewatering efficiency. However, there is currently a lack of rapid prediction technologies based on this molecular information.

[0004] Existing predictive models largely rely on traditional macroscopic physicochemical indicators and suffer from overfitting and weak generalization ability. While artificial neural networks have shown advantages in simulating complex systems, conventional backpropagation (BP) neural networks (such as LM-BP) are prone to prediction biases without manual parameter tuning, making it difficult to meet the accuracy and stability requirements of practical engineering. Therefore, developing a rapid prediction method for sludge dewatering performance based on key molecular indicators with high accuracy and strong stability has become an urgent need in the field of sludge treatment. Summary of the Invention

[0005] This application provides a rapid prediction method for the moisture content of sludge cake, which can solve the problem that prediction deviations are prone to occur under conditions without manual parameter adjustment, making it difficult to meet the accuracy and stability requirements of actual engineering projects.

[0006] To achieve the above objectives, this application provides a method for rapid prediction of sludge cake moisture content, the method comprising the following steps:

[0007] S1. Measure the specific resistance of sludge, the tyrosine content and tryptophan content in the conditioned dewatering filtrate and the moisture content of the filter cake after pressure filtration of several sludge samples to construct the original dataset; S2. After preprocessing the original dataset, the input and output components are imported into the BR-BP neural network for training. S3. Train the BR-BP neural network, adjust the parameters using the Bayesian regularization optimization algorithm, select the optimal parameters based on the evaluation index, and end the training to obtain the pre-trained BR-BP neural network. S4. Transmit the input components of the sludge to be predicted to the pre-trained BR-BP neural network and output the predicted value of the sludge cake moisture content.

[0008] Optionally, S1 further includes the following steps: Several sludge samples were collected, and conditioning and dewatering experiments were conducted using four different combinations of conditioning agents. The specific resistance of the sludge was measured using vacuum filtration, and the tyrosine and tryptophan contents of the corresponding batches of sludge conditioning and dewatering filtrate were measured using ultraviolet spectrophotometry. The moisture content of the sludge cake was calculated by the difference in sludge mass before and after filtration. The specific resistance of the sludge, the tyrosine content, the tryptophan content, and the moisture content of the sludge cake were used to construct an original dataset containing several sets of data.

[0009] Furthermore, the four conditioning agent combinations are respectively a combination of polyferric sulfate and hydrogen peroxide, a combination of polyferric sulfate and potassium persulfate, a combination of polyferric sulfate and sodium percarbonate, and a combination of polyferric sulfate and quaternary ammonium chitosan.

[0010] Optionally, step S2 further includes the following steps: The original dataset is subjected to min-max normalization and mapped to the interval [0,1] by linear transformation to obtain a normalized dataset. The sludge specific resistance, tyrosine content and tryptophan content of the normalized dataset are used as input components of the BR-BP neural network, and the sludge cake moisture content of the normalized dataset is used as the output component of the BR-BP neural network.

[0011] Furthermore, the BR-BP neural network adopts a three-layer feedforward topology, including an input layer, a single hidden layer, and an output layer. The input layer and the single hidden layer use the tansig activation function, and the output layer uses the purelin transfer function.

[0012] Furthermore, the Bayesian regularization optimization algorithm includes the following steps: Set the number of neurons in a single hidden layer to 8-15, the learning rate to 0.005-0.02, and the training precision to 1×10⁻⁶. -4 -1×10 -5 The BR-BP neural network is trained by comprehensively optimizing the objective function:

[0013] in, For fitting error, The number of training samples, This represents the actual moisture content of the mud cake. This is the predicted moisture content of the mud cake. As a weighted penalty term, To determine the severity of the punishment, The total number of network weights. For the sum of squares of the weights, For the first Each weight.

[0014] Furthermore, the evaluation indicators include the coefficient of determination, root mean square error, and mean absolute percentage error; the set of parameters whose coefficient of determination is closest to 1 and whose root mean square error and mean absolute percentage error are closest to 0 are selected as the optimal parameters.

[0015] Optionally, step S3 further includes the following steps: During each training iteration of the BR-BP neural network, the fitting error of the current iteration is calculated. and weighted penalty items ,pass Update penalty intensity ,in The effective weight number; the penalty intensity Constrained Dynamic adaptation error within the interval.

[0016] Furthermore, the fitting error and At that time, reduce the intensity of punishment. By relaxing penalties and adjusting network weights To improve the fit, the network weights for:

[0017] in, The learning rate; When considered as redundant weight, Considered as key weight, Then, the intensity of the punishment Adjusted to zero under constraints; When the fitting error and At that time, increase the intensity of punishment. To strengthen penalties and reduce redundant weights.

[0018] Optionally, S3 further includes the following steps: after each training iteration of the BR-BP neural network, the validation set error is calculated using the current model and the training parameters are saved; if the fitting error... If the error does not decrease, accumulate the number of consecutive iterations; if the error does not decrease after 6 consecutive iterations and the fitting error... Less than 1×10 -4When the error is reached, the early stopping strategy is triggered, terminating the training and saving the optimal parameters; if the fitting error is... If the number of consecutive iterations does not decrease, the training is stopped immediately when the number of iterations reaches 6. The saved optimal training parameters are then used as the optimal parameters to obtain the pre-trained BR-BP neural network.

[0019] In this application, sludge specific resistance, tyrosine content, and tryptophan content are used as input components to reflect sludge dewatering characteristics from both molecular and macroscopic levels, providing a more comprehensive and accurate prediction basis. A Bayesian regularization optimization algorithm is used to optimize the BP neural network, and the network weights are compressed by automatically adjusting the penalty intensity, effectively solving the problems of overfitting and weak generalization ability of conventional BP neural networks. High prediction accuracy and stability are maintained even without manual parameter tuning. This application only requires the measurement of three key indicators to achieve rapid prediction of cake moisture content, eliminating the need for complex simulation experiments, significantly shortening the detection cycle, reducing operating costs, and facilitating real-time application in engineering sites. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of a method for rapidly predicting the moisture content of sludge cake according to an embodiment of this application. Figure 2 This is a diagram of the BR-BP neural network structure provided according to an embodiment of this application; Figure 3 This is a comparison chart showing the reduction rate of sludge cake moisture content and tyrosine and tryptophan content in filtrate after treatment with "polyferric sulfate + hydrogen peroxide" according to the embodiments of this application. Figure 4 This is a comparison chart showing the reduction rate of sludge cake moisture content and tyrosine and tryptophan content in filtrate after treatment with "polyferric sulfate + potassium persulfate" according to the embodiments of this application. Figure 5 This is a comparison chart showing the reduction rate of sludge cake moisture content and tyrosine and tryptophan content in filtrate after treatment with "polyferric sulfate + quaternary ammonium salt chitosan" according to the embodiments of this application; Figure 6 This is a comparison chart showing the reduction rate of sludge cake moisture content and tyrosine and tryptophan content in filtrate after treatment with "polyferric sulfate + sodium percarbonate" according to the embodiments of this application; Figure 7 For the various prediction models provided according to the embodiments of this application Value comparison chart. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] Example 1 The purpose of this application is to overcome the shortcomings of existing technologies and provide a rapid prediction method for sludge cake moisture content, enabling rapid and accurate prediction of sludge dewatering performance and providing a scientific basis for the selection of sludge conditioning agents and process optimization. Figure 1 As shown, the rapid prediction method for sludge cake moisture content of this application includes the following steps: S1. Measure the specific resistance of sludge, the tyrosine content and tryptophan content in the conditioned dewatering filtrate and the moisture content of the filter cake after pressure filtration of several sludge samples to construct the original dataset; Several sludge samples were collected, and conditioning and dewatering experiments were conducted using four different combinations of conditioning agents. The specific resistance of the sludge was measured using vacuum filtration, and the tyrosine and tryptophan contents of the corresponding batches of sludge conditioning and dewatering filtrate were measured using ultraviolet spectrophotometry. The moisture content of the sludge cake was calculated by the difference in sludge mass before and after filtration. The specific resistance of the sludge, the tyrosine content, the tryptophan content, and the moisture content of the sludge cake were used to construct an original dataset containing several sets of data.

[0024] S2. After preprocessing the original dataset, the input and output components are imported into the BR-BP neural network for training. The original dataset is subjected to min-max normalization and mapped to the interval [0,1] by linear transformation to obtain a normalized dataset. The sludge specific resistance, tyrosine content and tryptophan content of the normalized dataset are used as input components of the BR-BP neural network, and the sludge cake moisture content of the normalized dataset is used as the output component of the BR-BP neural network.

[0025] S3. Train the BR-BP neural network, adjust the parameters using the Bayesian regularization optimization algorithm, select the optimal parameters based on the evaluation index, and end the training to obtain the pre-trained BR-BP neural network. During each training iteration of the BR-BP neural network, the fitting error of the current iteration is calculated. and weighted penalty items ,pass Update penalty intensity ,in The effective weight number; the penalty intensity Constrained Dynamic adaptation error within the interval; After each training iteration of the BR-BP neural network, the validation set error is calculated using the current model, and the training parameters are saved. If the fitting error... If the error does not decrease, accumulate the number of consecutive iterations; if the error does not decrease after 6 consecutive iterations and the fitting error... Less than 1×10 -4 When the error is reached, the early stopping strategy is triggered, terminating the training and saving the optimal parameters; if the fitting error is... If the number of consecutive iterations does not decrease, the training is stopped immediately when the number of iterations reaches 6. The saved optimal training parameters are then used as the optimal parameters to obtain the pre-trained BR-BP neural network.

[0026] S4. Transmit the input components of the sludge to be predicted to the pre-trained BR-BP neural network and output the predicted value of the sludge cake moisture content.

[0027] Example 2 In this embodiment, based on Example 1, the four conditioning agent combinations are respectively a combination of polyferric sulfate and hydrogen peroxide, a combination of polyferric sulfate and potassium persulfate, a combination of polyferric sulfate and sodium percarbonate, and a combination of polyferric sulfate and quaternary ammonium chitosan.

[0028] The BR-BP neural network adopts a three-layer feedforward topology, including an input layer, a single hidden layer, and an output layer. The input layer and the single hidden layer use the tansig activation function, and the output layer uses the purelin transfer function.

[0029] The Bayesian regularization optimization algorithm includes the following steps: Set the number of neurons in a single hidden layer to 8-15, the learning rate to 0.005-0.02, and the training precision to 1×10⁻⁶. -4 -1×10 -5 The BR-BP neural network is trained by comprehensively optimizing the objective function:

[0030] in, For fitting error, The number of training samples, This represents the actual moisture content of the mud cake. This is the predicted moisture content of the mud cake. As a weighted penalty term, To determine the severity of the punishment, The total number of network weights. For the sum of squares of the weights, For the first Each weight.

[0031] The evaluation indicators include the coefficient of determination, root mean square error, and mean absolute percentage error; the set of parameters whose coefficient of determination is closest to 1 and whose root mean square error and mean absolute percentage error are closest to 0 are selected as the optimal parameters.

[0032] The fitting error and At that time, reduce the intensity of punishment. By relaxing penalties and adjusting network weights To improve the fit, the network weights for:

[0033] in, The learning rate; When considered as redundant weight, Considered as key weight, Then, the intensity of the punishment Adjusted to zero under constraints; When the fitting error and At that time, increase the intensity of punishment. To strengthen penalties and reduce redundant weights.

[0034] Example 3 In one specific embodiment, eight batches of municipal sludge samples were collected from the secondary sedimentation tank of a wastewater treatment plant in a certain city. The specific resistance of the sludge to be dewatered was measured to be 1.32 × 10¹² m. kg - ¹ The filtrate contained 0.87 mg / L of tyrosine and 0.62 mg / L of tryptophan. The PFS+SPC conditioning scheme was adopted. The BR-BP neural network predicted that the moisture content of the filter cake was 55.32%, while the actual moisture content of the filter cake after pressure filtration was 55.27%, with a prediction accuracy of 99.91%.

[0035] In one specific embodiment, eight batches of municipal sludge samples were collected from the secondary sedimentation tank of a wastewater treatment plant in a certain city. The specific resistance of the sludge to be dewatered was measured to be 2.31 × 10¹² m. kg - ¹ The filtrate contained 1.24 mg / L of tyrosine and 0.95 mg / L of tryptophan. A polyferric sulfate + hydrogen peroxide conditioning scheme was used. The BR-BP neural network predicted that the moisture content of the cake was 62.13%, while the actual value was 62.05%, with a prediction accuracy of 99.87%.

[0036] Example 4 like Figure 3As shown, after conditioning with polyferric sulfate and hydrogen peroxide, the lowest moisture content of the mud cake reached 54.61%±0.8923%, with corresponding reduction rates of tyrosine and tryptophan of 70.4330%±2.9096% and 60.9057%±1.6971%, respectively, which was the highest among the eight treated batches. The highest moisture content of the mud cake reached 68.78%±0.5657%, with corresponding reduction rates of tyrosine and tryptophan of 64.3470%±3.4853% and 56.4990%±2.0711%, respectively, which was the lowest among the eight treated batches.

[0037] like Figure 4 As shown, after conditioning with polyferric sulfate and potassium persulfate, the lowest moisture content of the mud cake reached 53.64%±0.5297%, with corresponding reduction rates of tyrosine and tryptophan of 85.1698%±2.6167% and 70.6580%±3.8184%, respectively, which were the highest among the eight treated batches. The highest moisture content of the mud cake reached 62.74%±0.7823%, with corresponding reduction rates of tyrosine and tryptophan of 73.3319%±3.0711% and 60.6454%±4.8528%, respectively, which were the lowest among the eight treated batches.

[0038] like Figure 5 As shown, after conditioning with polyferric sulfate and quaternary ammonium salt chitosan, the lowest moisture content of the mud cake could reach 58.43%±1.5219%, with corresponding reduction rates of tyrosine and tryptophan of 73.3369%±3.2125% and 67.6811%±2.2225%, respectively, which were the highest among the eight treated batches. The highest moisture content of the mud cake reached 63.9%±1.4121%, with corresponding reduction rates of tyrosine and tryptophan of 69.6213%±2.8284% and 59.6120%±3.1113%, respectively, which were the lowest among the eight treated batches.

[0039] like Figure 6 As shown, after conditioning with polyferric sulfate and sodium percarbonate, the lowest moisture content of the mud cake could reach 50.8%±1.5297%, with corresponding reduction rates of tyrosine and tryptophan of 85.609%±3.8995% and 73.7338%±2.1213%, respectively, which were the highest among the eight treated batches. The highest moisture content of the mud cake reached 62.74%±1.7823%, with corresponding reduction rates of tyrosine and tryptophan of 73.3319%±3.0711% and 60.6455%±2.4853%, respectively, which were the lowest among the eight treated batches.

[0040] The results of the four experiments showed a consistent pattern after conditioning with the four combinations of agents: the moisture content of the sludge cake was significantly negatively correlated with the reduction rate of tyrosine and tryptophan in the filtrate. Specifically, the combination of polyferric sulfate and sodium percarbonate had the best conditioning effect, with the lowest sludge cake moisture content reaching 50.8%±1.5297%, and the corresponding reduction rates of tyrosine and tryptophan reaching as high as 85.609%±3.8995% and 73.7338%±2.1213%, respectively. After conditioning with polyferric sulfate and hydrogen peroxide, the highest sludge cake moisture content was 68.78%±0.5657%, with the lowest reduction rates of tyrosine and tryptophan, at 64.3470%±3.4853% and 56.4990%±2.0711%, respectively. This pattern fully verifies the scientific validity of aromatic proteins (tyrosine and tryptophan) in the extracellular polymeric substances of sludge as key molecular markers of dewatering performance. Their content changes directly reflect the sludge dewatering effect, providing core molecular-level input indicators for the prediction model.

[0041] Example 5 like Figure 7 As shown, the determination coefficients of five models were compared: a single-hidden-layer back-propagation (BP) feedforward neural network model based on the Leven–Marquardt (LM) algorithm (LM-BP model); a BP feedforward neural network model based on the Bayesian regularization (BR) algorithm (BR-BP model); a BP feedforward neural network model based on the Scaled Conjugate Gradient Method (SCG) algorithm (SCG-BP model); a double-hidden-layer feedforward neural network model also based on the Leven–Marquardt (LM) algorithm (ANN model); and an artificial neural network model based on the Random Forest (RF) algorithm (RF-ANN model). The determination coefficient of this application is the highest, reaching 0.877, closest to 1, indicating the best explanatory power for data variation. This superior performance is precisely due to the innovative design of this application. The innovative integration of multi-dimensional indicators breaks through the limitations of traditional models that rely solely on macroscopic indicators such as sludge specific resistance. It innovatively uses sludge specific resistance (a macroscopic physicochemical indicator) along with tyrosine and tryptophan content (molecular markers) as input components, achieving deep coupling between macroscopic dewatering characteristics and microscopic molecular mechanisms, laying a data foundation for improved prediction accuracy. This ensures the comprehensiveness and relevance of the input information. Furthermore, the Bayesian regularization algorithm addresses the overfitting and weak generalization capabilities of traditional BP networks, while the scientific network structure and training strategy guarantee the model's convergence efficiency and prediction stability. These results fully demonstrate that the BR-BP model has the best explanatory power for variations in sludge cake moisture content data, achieving high-precision and high-stability predictions without manual parameter tuning, meeting the needs of practical engineering applications.

[0042] Four combinations of polyferric sulfate-based composite conditioning agents effectively improved sludge dewatering performance, and the moisture content of the sludge cake showed a significant negative correlation with the tyrosine and tryptophan content in the filtrate, providing reliable indicators for the prediction model. The BR-BP prediction model based on Bayesian regularization optimization, through synergistic innovation in multi-dimensional indicator fusion, algorithm optimization, structural design, and training strategies, significantly outperformed mainstream comparative models, achieving an R² value as high as 0.877. This effectively solved the problems of large prediction bias and insufficient stability in existing technologies, providing an efficient and feasible technical solution for rapid and accurate prediction of sludge cake moisture content, and promoting the transformation of sludge treatment from experience-based control to data-driven approaches.

[0043] This application has the following beneficial effects: (1) This application uses tyrosine content, tryptophan content and sludge specific resistance as input components. These indicators are key factors affecting sludge dewatering performance and can reflect sludge dewatering characteristics from both molecular and macroscopic levels, making the prediction basis more comprehensive and accurate. (2) This application optimizes the BP neural network through the Bayesian regularization optimization layer and compresses the network weights by automatically adjusting the penalty intensity, which effectively solves the problems of easy overfitting and weak generalization ability of conventional BP neural networks. It can still maintain high prediction accuracy and stability under zero manual parameter tuning conditions. (3) The prediction process of this application is simple and fast. Only three key indicators need to be measured to achieve rapid prediction of the moisture content of the mud cake. No complicated simulation experiments are required, which greatly shortens the detection cycle, reduces operating costs, and facilitates real-time application in engineering sites. (4) This application is applicable to a variety of chemical conditioning process scenarios. The model is highly versatile and can provide a scientific basis for evaluating the sludge dewatering performance under different conditioning schemes, thus promoting the transformation of sludge treatment from experience-based control to data-driven approach.

[0044] This application also provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, they implement the various steps of the aforementioned method embodiment for rapid prediction of sludge cake moisture content, and can achieve the same beneficial effects as the aforementioned method embodiment for rapid prediction of sludge cake moisture content. To avoid repetition, they will not be described again here.

[0045] In the several embodiments provided in this application, it should be understood that the disclosed methods and apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between devices or units, and may be electrical, mechanical, or other forms.

[0046] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can be physically included separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0047] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute some steps of the transmission and reception methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0048] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principles described in this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for rapid prediction of sludge cake moisture content, characterized in that, The method includes the following steps: S1. Measure the specific resistance of sludge, the tyrosine content and tryptophan content in the conditioned dewatering filtrate and the moisture content of the filter cake after pressure filtration of several sludge samples to construct the original dataset; S2. After preprocessing the original dataset, the input and output components are imported into the BR-BP neural network for training. S3. The parameters are adjusted using the Bayesian regularization optimization algorithm. After selecting the optimal parameters based on the evaluation index, the training ends, and the pre-trained BR-BP neural network is obtained. S4. Transmit the input components of the sludge to be predicted to the pre-trained BR-BP neural network and output the predicted value of the sludge cake moisture content.

2. The method for rapid prediction of sludge cake moisture content according to claim 1, characterized in that, S1 further includes the following steps: Several sludge samples were collected, and conditioning and dewatering experiments were conducted using four different combinations of conditioning agents. The specific resistance of the sludge was measured using vacuum filtration, and the tyrosine and tryptophan contents of the corresponding batches of sludge conditioning and dewatering filtrate were measured using ultraviolet spectrophotometry. The moisture content of the sludge cake was calculated by the difference in sludge mass before and after filtration. The specific resistance of the sludge, the tyrosine content, the tryptophan content, and the moisture content of the sludge cake were used to construct an original dataset containing several sets of data.

3. The method for rapid prediction of sludge cake moisture content according to claim 2, characterized in that, The four combinations of conditioning agents are: a combination of polyferric sulfate and hydrogen peroxide, a combination of polyferric sulfate and potassium persulfate, a combination of polyferric sulfate and sodium percarbonate, and a combination of polyferric sulfate and quaternary ammonium chitosan.

4. The method for rapid prediction of sludge cake moisture content according to claim 3, characterized in that, S2 also includes the following steps: The original dataset is subjected to min-max normalization and mapped to the interval [0,1] by linear transformation to obtain a normalized dataset. The sludge specific resistance, tyrosine content and tryptophan content of the normalized dataset are used as input components of the BR-BP neural network, and the sludge cake moisture content of the normalized dataset is used as the output component of the BR-BP neural network.

5. The method for rapid prediction of sludge cake moisture content according to claim 4, characterized in that, The BR-BP neural network adopts a three-layer feedforward topology, including an input layer, a single hidden layer, and an output layer. The input layer and the single hidden layer use the tansig activation function, and the output layer uses the purelin transfer function.

6. The method for rapid prediction of sludge cake moisture content according to claim 5, characterized in that, The Bayesian regularization optimization algorithm includes the following steps: Set the number of neurons in a single hidden layer to 8-15, the learning rate to 0.005-0.02, and the training precision to 1×10⁻⁶. -4 -1×10 -5 The BR-BP neural network is trained by comprehensively optimizing the objective function: in, For fitting error, The number of training samples, This represents the actual moisture content of the mud cake. This is the predicted moisture content of the mud cake. As a weighted penalty term, To determine the severity of the punishment, The total number of network weights. For the sum of squares of the weights, For the first Each weight.

7. The method for rapid prediction of sludge cake moisture content according to claim 6, characterized in that, The evaluation indicators include the coefficient of determination, root mean square error, and mean absolute percentage error; the set of parameters whose coefficient of determination is closest to 1 and whose root mean square error and mean absolute percentage error are closest to 0 are selected as the optimal parameters.

8. The method for rapid prediction of sludge cake moisture content according to claim 7, characterized in that, S3 also includes the following steps: During each training iteration of the BR-BP neural network, the fitting error of the current iteration is calculated. and weighted penalty items ,pass Update penalty intensity ,in The effective weight number; the penalty intensity Constrained Dynamic adaptation error within the interval.

9. The method for rapid prediction of sludge cake moisture content according to claim 8, characterized in that, The fitting error and At that time, reduce the intensity of punishment. By relaxing penalties and adjusting network weights To improve the fit, the network weights for: in, The learning rate; When considered as redundant weight, Considered as key weight, Then, the intensity of the punishment Adjusted to zero under constraints; When the fitting error and At that time, increase the intensity of punishment. To strengthen penalties and reduce redundant weights.

10. The method for rapid prediction of sludge cake moisture content according to claim 9, characterized in that, S3 further includes the following steps: after each training iteration of the BR-BP neural network, the validation set error is calculated using the current model and the training parameters are saved; if the fitting error... If the number of consecutive occurrences does not decrease, then the cumulative number of occurrences is counted. When the error does not decrease after 6 consecutive iterations and the fitting error Less than 1×10 -4 When the error is reached, the early stopping strategy is triggered, terminating the training and saving the optimal parameters; if the fitting error is... If the number of consecutive iterations does not decrease, the training is stopped immediately when the number of iterations reaches 6. The saved optimal training parameters are then used as the optimal parameters to obtain the pre-trained BR-BP neural network.

Images