Sewage biochemical treatment dosing control system and method based on artificial intelligence

Abstract

The invention discloses a sewage biochemical treatment dosing control system and method based on artificial intelligence, and relates to the technical field of water treatment.The method comprises the steps that a state space is constructed, the fractal dimension of the state space is calculated, and a fractal evolution track is formed; determining a key state turning point through the spatial density of the abnormal fluctuation position; constructing a feature network taking the key state turning point as a node, and identifying a dominant state transition path from the feature network; on the basis of the dominant state transition path, determining a corresponding state triggering threshold condition, and when it is monitored that the state parameter data reach the state triggering threshold condition, triggering precise dispensing of the medicament at the corresponding position; after the medicament is put, the fractal dimension change is recalculated, and the selection condition of the key state turning point and the state triggering threshold value are dynamically adjusted; through state space fractal dimension dynamic analysis and self-adaptive feature network construction, accurate and efficient dynamic control of sewage treatment agent feeding is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water treatment, in particular to a sewage biochemical treatment dosing control system and method based on artificial intelligence. BACKGROUND

[0002] As a mature sewage biochemical treatment technology, activated sludge process is widely used in the fields of pollution control of municipal sewage and industrial wastewater. The sewage biochemical treatment process is essentially a highly complex nonlinear dynamic process, in which multiple biochemical reactions are coupled, and various state parameters (such as dissolved oxygen concentration, oxidation-reduction potential, pH value, chemical oxygen demand, etc.) show obvious non-stationarity and multi-scale spatial heterogeneity characteristics. Due to the complex dynamic relationship between these state parameters, the traditional monitoring method based on single or static parameter analysis often fails to accurately reveal the change rule of the real state in the sewage treatment process, and cannot timely identify and locate the abnormal state occurrence position and evolution trend.

[0003] At the same time, in actual sewage treatment system, the occurrence of abnormal working condition is often accompanied by significant reduction of sewage treatment efficiency or deterioration of sludge performance, and the existing technology has problems of slow response, high false alarm rate and fuzzy space-time positioning in state monitoring and abnormal identification, which is difficult to realize accurate identification and efficient early warning of abnormal working condition. In addition, the dosing strategy of traditional sewage treatment system is usually to set the amount, location and time point of dosing based on fixed timing or artificial experience. This method not only ignores the dynamic change characteristics of the state of sewage treatment system, but also may cause excessive or insufficient dosing of reagent, making it difficult to achieve precise control of reagent dosing, seriously affecting the operation efficiency of sewage treatment and increasing the operation cost. SUMMARY

[0004] The purpose of the present application is to provide a sewage biochemical treatment dosing control system and method based on artificial intelligence to solve the problems in the background art.

[0005] In order to achieve the above purpose, the present application provides the following technical scheme:

[0006] In the first aspect, the present application provides a sewage biochemical treatment dosing control method based on artificial intelligence, comprising:

[0007] S101: acquiring state parameter data continuously collected at multiple points in a to-be-treated area, constructing a state space according to the correlation between parameters, calculating the fractal dimension of the state space, and forming a fractal evolution trajectory of the state space;

[0008] S102: identifying multiple abnormal fluctuation positions based on the morphological features of the fractal evolution trajectory, and determining a key state turning point through the spatial density of the abnormal fluctuation positions;

[0009] S103: constructing a feature network with the key state turning points as nodes according to the distribution mode of the key state turning points, and identifying a dominant state transition path from the feature network;

[0010] S104: determining a corresponding state trigger threshold condition based on the dominant state transition path, and monitoring state space data changes in real time, and triggering precise dispensing of medicaments at a corresponding position when it is monitored that state parameter data reaches the state trigger threshold condition;

[0011] S105: after the medicaments are dispensed, recalculating the fractal dimension change of the state space in the to-be-processed region, and dynamically adjusting the selection condition of the key state turning points and the state trigger threshold, so as to realize dynamic adaptive control of the precise dispensing of medicaments.

[0012] In a second aspect, the present application provides a sewage biochemical treatment dosing control system based on artificial intelligence, which is realized based on the sewage biochemical treatment dosing control method based on artificial intelligence described above, and comprises:

[0013] A trajectory construction module is configured to acquire state parameter data continuously collected at multiple points in a to-be-processed region, construct a state space according to the correlation between parameters, calculate the fractal dimension of the state space, and form a fractal evolution trajectory of the state space;

[0014] An abnormal point identification module is configured to identify multiple abnormal fluctuation positions based on the morphological features of the fractal evolution trajectory, and determine key state turning points through the spatial density of the abnormal fluctuation positions;

[0015] A network construction and path searching module is configured to construct a feature network with the key state turning points as nodes according to the distribution mode of the key state turning points, and identify a dominant state transition path from the feature network;

[0016] A threshold control and dispensing module is configured to determine a corresponding state trigger threshold condition based on the dominant state transition path, and monitor state space data changes in real time, and trigger precise dispensing of medicaments at a corresponding position when it is monitored that state parameter data reaches the state trigger threshold condition;

[0017] A self-adjusting closed loop module is configured to, after medicaments are dispensed, recalculate the fractal dimension change of the state space in the to-be-processed region, and dynamically adjust the selection condition of the key state turning points and the state trigger threshold, so as to realize dynamic adaptive control of the precise dispensing of medicaments.

[0018] In the above technical solution, the present application provides the following technical effects and advantages:

[0019] The application can comprehensively and accurately represent the non-stationary dynamic change characteristics of the internal state parameters of the sewage treatment system by introducing the dynamic correlation structure of the state space and the fractal dimension evolution trajectory in the sewage treatment process state monitoring; effectively improves the timeliness, accuracy and spatial positioning accuracy of state abnormal fluctuation position identification; realizes real-time monitoring and early warning of the abnormal state of the sewage treatment process, thereby significantly improving the stability and reliability of the sewage treatment process operation.

[0020] The application accurately determines the position and dynamic evolution law of the key state turning point in the sewage treatment system by adopting the abnormal fluctuation position clustering analysis method based on the two-dimensional characteristics of spatial density and spatial distribution non-uniformity; further constructs an optimized feature network structure to accurately identify the dominant transition path between states; realizes accurate extraction of the key state turning point and efficient analysis of the state transition process, and effectively improves the monitoring ability of the sewage treatment process to abnormal state.

[0021] The application can real-time monitor and accurately respond to the change trend of the state parameters in the sewage treatment process by designing a dynamic adaptive drug accurate injection strategy based on the dominant state transition path; by dynamically adjusting the key state turning point discrimination threshold and state triggering threshold, the accurate optimization control of the drug injection time, position and dosage is realized; effectively reduces the blindness and waste of drug injection, reduces the operation cost of the sewage treatment process, and significantly improves the stability and sustainability of the sewage treatment effect. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art according to these drawings.

[0023] Figure 1 The flow chart of the present application is a sewage biochemical treatment dosing control method based on artificial intelligence.

[0024] Figure 2 The framework diagram of the present application is a sewage biochemical treatment dosing control system based on artificial intelligence. DETAILED DESCRIPTION

[0025] Example implementations are now described with reference to the drawings. Example implementations can, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these example implementations are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the example implementations to those skilled in the art. Undepicted features, structures or characteristics can be combined in any suitable manner in one or more example implementations. Numerous specific details are provided to provide a thorough understanding of the example implementations of the present disclosure. One skilled in the relevant art will recognize, however, that the techniques of the present disclosure can be practiced without one or more of the specific details, or with other methods, components, materials, and so forth. In other instances, well-known structures, materials, or operations are not shown or described in detail in order to avoid obscuring aspects of the present disclosure.

[0026] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more example implementations. In the following description, numerous specific details are provided to give a thorough understanding of example implementations of the present disclosure. One skilled in the relevant art will recognize, however, that the techniques of the present disclosure can be practiced without one or more of the specific details, or with other methods, components, materials, and so forth. In other instances, well-known structures, materials, or operations are not shown or described in detail in order to avoid obscuring aspects of the present disclosure.

[0027] Example 1

[0028] As Figure 1 shown, the present embodiment discloses a sewage biochemical treatment dosing control method based on artificial intelligence, comprising:

[0029] S101: Obtain state parameter data continuously collected at multiple points in a to-be-processed area, construct a state space according to the correlation between parameters, calculate the fractal dimension of the state space, and form a fractal evolution trajectory of the state space;

[0030] The to-be-processed area of the present embodiment is a biochemical reaction tank for sewage treatment using the activated sludge method, and the state parameter data specifically includes dissolved oxygen concentration (DO), oxidation-reduction potential (ORP), pH value, and chemical oxygen demand (COD). The state parameter data is collected in real time by a multi-point sensor array arranged inside the reaction tank and transmitted to a background processing device;

[0031] Exemplarily, a plurality of sensor arrays are arranged inside the biochemical reaction tank, each sensor array is uniformly arranged along the direction of sewage flow, and each sensor array includes a plurality of probes for collecting the DO, ORP, pH, and COD values of a plurality of spatial points in the tank in real time;

[0032] In a specific implementation, the construction of the state space, the calculation of the fractal dimension of the state space, and the formation of the fractal evolution trajectory of the state space include:

[0033] According to the collection frequency and spatial distribution characteristics of the parameters in the to-be-processed area, the length of the space-time coupling window of the state parameters is determined;

[0034] It should be understood that, due to the difference in the collection frequency of different sensors and the regularity of the spatial distribution of the sensors inside the biochemical reaction tank along the sewage flow direction, in order to ensure that the data of each state parameter can be analyzed in a unified time-space scale, the embodiment needs to determine a unified time-space coupling window length;

[0035] It should be noted that the specific method for determining the unified time window length is as follows: first, the lowest frequency value f min in all sensor collection frequencies is counted, and then an integer multiple of the sampling period of the lowest frequency is taken as the time window length: In the formula, T window is the unified time window length, and n is an integer multiple, which can be adaptively adjusted according to the field data;

[0036] Further, the specific method for determining the unified spatial window length is as follows: first, the distance L s between adjacent sensor arrays along the sewage flow direction is obtained, and then an integer multiple of the distance is taken as the spatial window length: L window = m x L s , m is an integer multiple, and the specific value is determined according to the length of the tank body and the process reaction characteristics;

[0037] It can be understood that the time window length and the spatial window length determined by the above method together form a time-space coupling window, thereby providing a unified time-space analysis scale for subsequent state space construction.

[0038] The time-domain fluctuation characteristics and the spatial distribution difference characteristics of the state parameters in the time-space coupling window are extracted to form a multi-scale feature set;

[0039] Specifically, the time-domain fluctuation characteristics are specifically the fluctuation amplitude and the change rate of the data sequence of each state parameter in the time-space coupling window, and the spatial distribution difference characteristics are specifically the variance value between the parameter values of each measurement point in the spatial window at the same time;

[0040] The formula for calculating the time-domain fluctuation characteristics is as follows:

[0041]

[0042] In the formula, A var is the fluctuation amplitude of the parameter in the time window; x t is the value of the parameter at time t in the time window; is the average value of the parameter sequence in the window; and N is the number of data points in the window;

[0043] The formula for calculating the spatial distribution difference characteristics is as follows:

[0044]

[0045] In the formula, S diff (t) is a parameter of the spatial distribution difference feature of the parameter in the spatial window; x i (t) is a parameter value of the ith measurement point at time t; is the average value of all measurement points in the spatial window at time t; and M is the total number of measurement points in the spatial window;

[0046] The two types of features above jointly form a multi-scale feature set to sufficiently represent the dynamic changes of the state parameters in the time domain and the spatial domain.

[0047] Based on the multi-scale feature set, a non-stationary correlation index between the state parameters is calculated to determine the dynamic association structure of the state parameters.

[0048] It should be understood that, since the sewage biochemical treatment process is a non-stationary dynamic process, the association between the state parameters can change significantly over time, and therefore an index capable of describing the non-stationary correlation between the parameters is required.

[0049] Specifically, the determination of the dynamic association structure of the state parameters comprises:

[0050] A sliding window cross-spectrum analysis method is used to extract the time-varying frequency domain coherence between different state parameters.

[0051] Exemplarily, taking the DO and ORP two state parameters as examples, first, a data sequence is gradually intercepted by using a sliding window with a length of L seg and a step of L step in the determined spatio-temporal coupling window, and the cross power spectrum density function P xy (f,t) and the respective self power spectrum density functions P xx (f,t), P yy (f,t) in each window are calculated, and then a time-varying frequency domain coherence index in each window is calculated:

[0052]

[0053] In the formula, C xy (f,t) is the coherence of the parameter x and the parameter y at frequency f and time t.

[0054] Similarly, the time-varying frequency domain coherence between all pairs of state parameters is calculated.

[0055] Based on the time-varying frequency domain coherence, a time-frequency coherence spectrum is generated, and a high coherence region in the spectrum is extracted.

[0056] It should be noted that the time-frequency coherence spectrum is a time-frequency representation of the aforementioned time-varying frequency domain coherence index C xy(f, t) time and frequency two-dimensional spectrum formed by drawing; in the spectrum, further extract the region where the coherence value is greater than a preset threshold C th as a high coherence region.

[0057] According to the duration and intensity of the high coherence region, the significance weight of the parameter correlation in the dynamic correlation structure is determined.

[0058] It should be understood that the duration and coherence intensity of the high coherence region can jointly reflect the stability and strength of the dynamic correlation between parameters, and therefore the calculation formula of the significance weight is defined as follows:

[0059]

[0060] In the formula, W xy is the significance weight of parameters x and y; T total is the total duration of the analysis period; t k is the duration of the kth high coherence region; C avg,k is the average intensity of the coherence in the kth high coherence region.

[0061] The significance weight of all parameter pairs is determined in this way.

[0062] According to the significance weight, the state space parameters are optimized to represent the correlation between the state parameters in the dynamic correlation structure.

[0063] In specific implementation, it should be noted that according to the significance weight, the parameter pair with a weight greater than a weight threshold W th is selected, and the corresponding parameters are taken as the characteristic parameters of the state space. After optimization in this way, the parameter combination with the most significant dynamic correlation is retained to construct the state space, and the dynamic characteristics of the correlation structure between the state parameters are embodied.

[0064] According to the dynamic correlation structure, the state space is constructed, the fractal dimension is calculated based on the irregularity of the parameter trajectory in the state space, and the fractal evolution trajectory of the state space is obtained.

[0065] It should be understood that the state space is constructed by taking the state parameters determined in the dynamic correlation structure as the coordinates of each dimension of the space, and the real-time acquisition values of the state parameters in the time-space coupling window constitute the trajectory of the multi-dimensional state space.

[0066] Further, the box dimension method is used to calculate the fractal dimension of the state space trajectory to quantify the irregularity of the trajectory.

[0067] The box dimension calculation formula is as follows:

[0068]

[0069] wherein: D is the box dimension; ∈ is the box size covering the trajectory of the state space; N(∈) is the number of boxes required to cover the trajectory of the state space.

[0070] S102: Identify a plurality of abnormal fluctuation positions based on the morphological features of the fractal evolution trajectory, and determine a key state turning point by the spatial density of the abnormal fluctuation positions.

[0071] In a specific implementation, the determination of the key state turning point comprises:

[0072] Identify the abnormal fluctuation positions based on the duration and amplitude of the local mutation of the fractal evolution trajectory.

[0073] It should be noted that the fractal evolution trajectory in this embodiment is a curve of the fractal dimension changing with time, which usually presents the characteristics of local stability and local severe fluctuation alternately; in order to accurately identify the abnormal fluctuation positions, a sliding window length for abnormal fluctuation identification needs to be determined first.

[0074] Specifically, a sliding window with a length of L diff is set, which is scanned along the fractal evolution trajectory gradually, and the local difference value of the data in the window is calculated, which is as follows:

[0075]

[0076] wherein: D diff (t) is the local difference value at time t; D th and D min are the fractal dimension values corresponding to the front and rear endpoints of the window respectively.

[0077] Further, when the local difference value is greater than a preset difference threshold D th and the duration is greater than a minimum duration threshold T min , the corresponding position is marked as an initial abnormal fluctuation position.

[0078] The spatial local density and the spatial distribution unevenness index are calculated for the local mutation characteristics of the abnormal fluctuation positions.

[0079] It should be understood that since the abnormal fluctuation positions may occur at different positions in the reaction cell space, in order to determine the spatial density, the spatial feature analysis of the abnormal fluctuation positions is required.

[0080] In a specific implementation, the spatial local density is calculated as follows: ρ i is the spatial local density of the i-th initial abnormal fluctuation position; n i is the number of abnormal fluctuation positions in the spatial spherical neighborhood with the i-th initial abnormal fluctuation position as the center and a radius of r; V r is the spatial volume of the sphere with a radius of r.

[0081] Further, the spatial distribution unevenness index is defined as the variance of the distance from the abnormal position to other positions in its spatial neighborhood: U i is the spatial distribution unevenness index; d ij is the Euclidean distance between the ith abnormal fluctuation position and the jth position in its spatial neighborhood; is the average distance from the ith position to other positions in its spatial neighborhood.

[0082] The spatial local density and the spatial distribution unevenness index are used for two-dimensional clustering to obtain high-density areas of abnormal fluctuation positions;

[0083] It should be noted that the two-dimensional clustering method based on density and unevenness is used to cluster and analyze the abnormal fluctuation positions in the embodiment to determine the high-density areas of spatial distribution;

[0084] In the specific implementation process, first, all abnormal fluctuation positions are represented in a two-dimensional feature space of spatial local density ρ i and spatial distribution unevenness index U i , wherein each abnormal position corresponds to a feature point (ρ i , U i );

[0085] The DBSCAN density clustering algorithm is preferably used to cluster and analyze the above feature points in the embodiment, and the clustering result of the feature points is determined by two key parameters: neighborhood radius ε and minimum number of sample points in the neighborhood MinPts;

[0086] In the specific implementation, first, the number of points within a distance of ε from each feature point in the two-dimensional feature space is calculated, and when the number of points in the neighborhood of the point reaches or exceeds MinPts, the point is defined as a core point, and all points within a distance of less than ε from the core point form a cluster (high-density area);

[0087] Further, the neighborhood radius ε in the embodiment can be determined by the spatial scale and distribution of the abnormal fluctuation positions, and is generally 1-2 times the average Euclidean distance between the feature points of the abnormal positions; the minimum number of sample points in the neighborhood MinPts is usually 3-5 to effectively identify areas with higher spatial density;

[0088] After the above clustering, a number of high-density areas of abnormal fluctuation positions are formed, and each high-density area corresponds to a region with a significant spatial density.

[0089] The key state turning points are determined according to the continuity and stability of the high-density areas and the clear degree of boundaries between regions;

[0090] It should be understood that there may be differences in the persistence stability and the clearness of the boundaries between regions in the high-density area of spatial anomaly fluctuation, and the embodiment further determines the key state transition point by evaluating these differences;

[0091] Specifically, for each high-density area, the persistence stability of the high-density area is defined as the frequency of repeated occurrence of the position of the anomaly fluctuation in the high-density area in multiple continuous spatiotemporal coupling windows, and the specific calculation formula is as follows: In the formula, R j is the persistence stability of the jth high-density area, N rep,j is the number of times of repeated occurrence of the position of the anomaly fluctuation in the high-density area in multiple continuous spatiotemporal coupling windows; N total is the total number of spatiotemporal coupling windows in the analysis period.

[0092] In addition, the clearness of the boundaries between regions is defined as the average spatial distance between the high-density area and the adjacent high-density area: In the formula, B j is the clearness of the boundaries of the jth high-density area; K j is the number of other high-density areas adjacent to the jth high-density area; D jk is the Euclidean distance between the center points of the jth high-density area and the kth adjacent high-density area.

[0093] It should be noted that the embodiment sets the persistence stability threshold R th and the boundary clearness threshold B th When a certain high-density area meets R j ≥ R th and B j ≥ B th , the center position of the region is determined as the key state transition point.

[0094] Specifically, the center position of the region is determined by the average value of the spatial coordinates of all anomaly positions in the high-density area, so that the position of the key state transition point is clearly and accurately determined.

[0095] S103: According to the distribution mode of the key state transition point, a feature network with the key state transition point as a node is constructed, and a dominant state transition path is identified from the feature network.

[0096] In specific implementation, the construction of the feature network with the key state transition point as a node and the identification of the dominant state transition path from the feature network include:

[0097] The transmission probability and time delay characteristics of the key state transition point in the state space trajectory are calculated.

[0098] The key state transition point in the embodiment represents a position where the sewage treatment state is transformed from one relatively stable state to another relatively stable state. To clarify the correlation between the key transition points, the transmission probability and time delay between the key state transition points need to be counted first.

[0099] Specifically, assume that the state space trajectory is located at the key transition point v i at time t i , and then is transferred to another key transition point v j at time t j . The transmission probability from the key transition point v i to the key transition point v j is defined as follows: Further, the time delay feature between the key transition points is defined as follows: T ij represents the average time delay from the key transition point v i to the key transition point v j ; t j，m and t i，m respectively represent the time when the mth occurrence of the transition point v i and the transition point v j immediately following it occur.

[0100] It can be understood that the transmission probability and the time delay feature together depict the time sequence dependence between the key state transition points, providing sufficient basis for constructing the subsequent feature network.

[0101] Based on the transmission probability and the time delay feature, an initial correlation network between the transition points is constructed, and the node degree and the clustering coefficient are used to represent the node importance.

[0102] In the implementation, the embodiment first constructs a directed and weighted network graph G based on all the key state transition points. The node set V = {v1, v2,..., v n} of the network is all the key state transition points, the edge set E represents the transition point pairs that have a direct transition relationship, and the edge weight W = {w ij} represents the correlation strength between the transition points.

[0103] It should be noted that the weight w ij of the edge in the embodiment is determined by integrating the transmission probability and the time delay feature, and the specific calculation formula is as follows: where T ij is the corresponding average time delay, and T0 is a time delay normalization parameter, which generally takes the average value of all the time delay features.

[0104] Further, the node importance is comprehensively measured by two indexes of node degree and clustering coefficient:

[0105] The node degree is defined as the sum of in-degree and out-degree of the critical state transition point:

[0106] The clustering coefficient is defined as the degree of mutual connection between adjacent nodes of the critical transition point: Wherein, E i is the actual number of connections between all adjacent nodes of node v i ; k i is the total number of adjacent nodes of node v i .

[0107] Through the above two indexes, the embodiment can preliminarily determine the importance degree of the critical transition point in the network structure.

[0108] Based on the node importance index, the initial associated network is weighted, and a link prediction method is used to optimize the feature network structure.

[0109] Specifically, the link prediction method used to optimize the feature network structure includes:

[0110] An initial link feature matrix is constructed based on the historical transfer records between state transition points.

[0111] In specific implementation, assuming that the total number of critical transition points is n, an initial link feature matrix L ∈ R n×n is defined, and the matrix element is defined as:

[0112]

[0113] It can be understood that the matrix clearly records the frequency of historical state transition and can reflect the initial association strength between nodes in the network.

[0114] The probability distribution of potential connection between nodes is obtained by using a random walk algorithm, and the predicted connection strength is calculated in combination with the link feature matrix.

[0115] Exemplarily, in the embodiment, a classical random walk algorithm (Random Walk with Restart, RWR) is preferably used to predict the potential connection relationship between nodes in the feature network structure, and the specific implementation process is as follows:

[0116] First, a probability transition matrix M of the network is constructed, and the matrix element is defined as: Further, the stable transition probability R (t+1) between each pair of nodes in the network is calculated by using the random walk algorithm: (t) R (0) = (1-α)MR + αR , wherein R(t) represents the visiting probability vector of each node after the t-th random walk; a is a restart probability parameter, generally ranging from 0.1 to 0.3; R (0) is an initial visiting probability vector, which can be uniformly distributed or distributed according to the importance weight of the node;

[0117] After multiple iterations until stable, the stable probability distribution of the potential connection between nodes is obtained, and the stable probability distribution matrix R is defined, and the matrix element R ij represents the long-term stable probability that the random walk starting from the node v i arrives at the node v j ; further, the predicted connection strength between the node pairs is determined according to the stable probability matrix R and the initial link feature matrix L: wherein max(R), max(L) respectively represent the maximum value in the matrix R, L, for normalization.

[0118] According to the predicted connection strength, the link of the initial associated network between nodes is enhanced or inhibited, and the network structure is optimized;

[0119] Specifically, the embodiment sets a connection strength threshold S th , if the predicted connection strength S ij >S th , and the initial network does not exist, a new link is added to the network; if S ij <S' th is a lower threshold, S' th <S th ) and the initial network exists, the link is removed from the network;

[0120] It should be noted that the above threshold S th , S' th is determined according to the statistical characteristics of the historical network structure, for example, S th can be set as the 75% quantile value of the predicted connection strength, and S' th can be set as the 25% quantile value of the predicted connection strength.

[0121] According to the difference of path connectivity before and after the network structure optimization, the optimization effect is evaluated, and the final feature network structure is determined;

[0122] Specifically, the evaluation of the optimization effect includes:

[0123] According to the dominant state transition path before and after the feature network structure optimization, the coverage degree of the path to the abnormal fluctuation position is calculated;

[0124] It can be understood that the dominant state transition path is specifically defined as a path in the feature network with an edge weight (connection strength) greater than a threshold; the embodiment preferably uses a path search algorithm (such as Dijkstra algorithm) to identify the path with the highest weight between nodes, and counts the proportion of abnormal fluctuation positions corresponding to the key turning points in the state space passed by the path;

[0125] Determine the change in accuracy of the feature network in identifying key state turning points after the coverage is improved;

[0126] Further, the accuracy change is defined as the difference between the proportion of path coverage of abnormal fluctuation positions after optimization and the proportion of path coverage of abnormal fluctuation positions before optimization: Where: ΔA cover is the coverage accuracy change; is the proportion of the path coverage of abnormal fluctuation positions after optimization and before optimization, respectively;

[0127] Based on the change in recognition accuracy, the correlation between the accuracy of the drug delivery and the optimization and adjustment of the key state turning points is quantified.

[0128] In specific implementation, the correlation is defined as the correlation coefficient between the change in recognition accuracy of the feature network before and after optimization and the improvement in drug delivery accuracy, which can be calculated using the Pearson correlation coefficient.

[0129] According to the correlation, the overall improvement effect of the feature network optimization and adjustment on the state transition monitoring effect is evaluated, and the optimization effect is evaluated.

[0130] For example, when the correlation is higher than a preset threshold (such as 0.7), it indicates that the feature network structure after optimization can significantly improve the accuracy of key state recognition and the accuracy of drug delivery, thereby finally confirming the optimized feature network structure.

[0131] According to the optimized feature network structure, the stability and uniqueness of the associated path between nodes are calculated, and the dominant state transition path is identified.

[0132] It should be understood that the stability of the dominant state transition path is defined as the continuous key node transition probability in the path, and the uniqueness is defined as the degree of non-competition of the path in the network (i.e. the inverse of the proportion of repeated nodes on the path and other paths) ;

[0133] In specific implementation, the path stability and uniqueness are calculated as follows:

[0134] Path stability:

[0135]

[0136] Path uniqueness:

[0137]

[0138] Finally, the path stability and uniqueness are weighted and combined to identify the comprehensive optimal dominant state transition path, which explicitly guides the precise trigger conditions for subsequent drug injection.

[0139] S104: Based on the dominant state transition path, determine the corresponding state trigger threshold condition, and monitor the state space data in real time. When the state parameter data reaches the state trigger threshold condition, trigger the precise injection of the corresponding position drug;

[0140] In specific implementation, the determination of the corresponding state trigger threshold condition comprises:

[0141] According to the state parameter distribution characteristics of the nodes in the dominant state transition path, the extreme value and the change rate of the state parameter are counted.

[0142] Specifically, the embodiment first extracts the time position corresponding to each node (i.e. key state turning point) and the historical data of each state parameter recorded in the state space from the dominant state transition path; The state parameters include dissolved oxygen concentration (DO), oxidation-reduction potential (ORP), pH value and chemical oxygen demand (COD).

[0143] Exemplarily, taking DO as an example, the specific implementation method is as follows:

[0144] (1) For each key node v i in the dominant state transition path, the time sequence of DO value is extracted from the historical trajectory data of the state space, denoted as DO i (t), where t is the time interval around the node v i ;

[0145] (2) Calculate the extreme value (i.e. maximum and minimum) of DO parameter of each node, the specific formula is as follows:

[0146]

[0147] Where T i represents the specific analysis period corresponding to the key node v i ;

[0148] (3) Further calculate the change rate of DO parameter at each node, using the first-order difference method, the specific calculation formula is:

[0149]

[0150] Where V DO，i (t) is the node v iThe instantaneous change rate of DO concentration at the surrounding time t; Δt is the differential step length used for calculating the rate, which is determined according to the state parameter sampling frequency, and the typical value ranges from 1 minute to 5 minutes;

[0151] The extreme value and change rate of all nodes and other state parameters (ORP, pH, COD) on the path are calculated in the same way to obtain a comprehensive and clear set of state parameter change characteristics.

[0152] According to the extreme value and change rate of the state parameters, an initial state triggering threshold interval is constructed, and the early warning sensitivity of each threshold interval is determined;

[0153] It should be noted that in the embodiment, the change rate threshold V th，i According to the statistical characteristics of the historical data change rate, the typical selected value is the 75% to 90% quantile value of the historical change rate, so as to ensure the rationality of the sensitivity;

[0154] Through the above calculation, a series of initial state triggering threshold intervals and corresponding early warning sensitivity data are formed for each state parameter, thereby providing a clear initial basis for further optimization of the threshold interval.

[0155] According to the historical coincidence degree between the early warning sensitivity and the actual state transition event, a multi-objective optimization method is used to dynamically adjust the threshold interval;

[0156] It should be noted that in the embodiment, the early warning sensitivity S warn,i The coincidence degree between the actual occurrence of an abnormal state transition event is used to determine the rationality of the initial threshold interval;

[0157] The specific implementation is:

[0158] (1) The coincidence degree index is defined as the accuracy rate (Precision) and recall rate (Recall) of the early warning signal correctly capturing the actual abnormal event, and the specific calculation method is:

[0159]

[0160] Wherein: TP i (True positive) indicates the number of times that the node v i The early warning signal correctly captures the actual state transition event; FP i (False positive) indicates the number of times that the node v i The early warning signal is false; FN i (False negative) indicates the number of times that the node v i The early warning signal fails to capture the actual state transition event;

[0161] (2) further utilize multi-objective optimization method (such as genetic algorithm or particle swarm algorithm) to dynamically adjust the initial state triggering threshold interval, so as to optimize the above-mentioned precision and recall indicators at the same time;

[0162] Specifically, the objective function of the multi-objective optimization method is defined as:

[0163] MaximizeF=w1×Precision i +w2×Recall i

[0164] In the formula, the weight coefficients w1 and w2 are determined according to experimental data;

[0165] After optimization and adjustment, the final dynamic state triggering threshold interval of each state parameter is obtained, which ensures that the threshold setting reaches a good balance between recognition accuracy and early warning timeliness.

[0166] When the state parameter change rate in the dynamically adjusted threshold interval meets the preset triggering condition, the state triggering threshold condition is determined and the medicament is triggered;

[0167] In specific implementation, the embodiment defines the real-time monitoring triggering condition as follows:

[0168] (1) Real-time monitoring of state parameters at each measurement point in the biochemical reaction pool, and real-time calculation of the change rate;

[0169] (2) When the instantaneous value of the real-time state parameter enters the dynamically adjusted triggering threshold interval, and the parameter change rate exceeds the dynamically determined change rate triggering threshold, a medicament dispensing triggering signal is generated;

[0170] (3) After the triggering signal is sent out, the position and dose of the medicament accurate dispensing are determined through the preset control strategy combined with the abnormal fluctuation position and the corresponding key state turning point space region, and the control signal is sent to the corresponding medicament dispensing execution device to complete the accurate dispensing operation;

[0171] The above process ensures that the accurate dispensing of the medicament can timely respond to the state change trend, and realizes effective control of the abnormal state of the sewage treatment.

[0172] S105: After the medicament is dispensed, the fractal dimension change of the state space in the to-be-processed region is recalculated, and the selection condition of the key state turning point and the state triggering threshold are dynamically adjusted to realize dynamic adaptive control of the accurate dispensing of the medicament;

[0173] In specific implementation, the dynamic adjustment of the selection condition of the key state turning point includes:

[0174] Based on the state space fractal dimension change curve after the drug is put, the time delay characteristics and the space diffusion characteristics of the drug effect are identified;

[0175] In specific implementation, the state parameter data is continuously collected after the drug is accurately put, and the fractal dimension change curve of the state space trajectory is recalculated by using the same method in step S101, and the change characteristics of the state space under the action of the drug are observed, specifically as follows:

[0176] The time delay characteristic index T of the drug effect is defined lag The time difference between the time when the state space fractal dimension first changes significantly (greater than the preset threshold) after the drug is put and the time when the drug is put, and the specific formula is as follows: T lag =t change -t inject , wherein t change is the time when the state space fractal dimension first changes significantly; t inject is the time when the drug is actually put;

[0177] (2) The space diffusion characteristic index L of the drug is defined spread The range of the region where the fractal dimension change curve in the space window reaches stability after the drug is put is determined by gradually analyzing the change rate of the fractal dimension at each spatial position along the spatial axis;

[0178] It can be understood that the time delay characteristics and the space diffusion characteristics together depict the dynamic characteristics of the drug effect, and provide a basis for subsequent adjustment of the key state turning points and the trigger threshold.

[0179] According to the time delay characteristics and the space diffusion characteristics of the drug effect, the spatial density criterion in abnormal fluctuation identification is adjusted;

[0180] In specific implementation, the embodiment considers that the space diffusion characteristics of the drug have a significant impact on the density of the abnormal fluctuation position, therefore, the spatial density criterion for abnormal fluctuation identification needs to be dynamically adjusted after the drug effect, specifically as follows:

[0181] The initial spatial local density criterion is adjusted after the drug effect, and the specific method is as follows:

[0182]

[0183] In the formula, r adjust , MinPts adjust are the adjusted neighborhood radius and the minimum number of sample points respectively; L base , T base are the space diffusion and time delay characteristic reference values in the standard state respectively, and γ, δ are adjustment factors, and the typical value range is 0.1-0.5, which is determined by self-adaptive optimization according to historical data.

[0184] Through the above adjustment, the spatial clustering analysis of the abnormal fluctuation positions after the action of the medicament can be ensured to be more in line with the actual situation, thereby improving the identification accuracy of the key state turning point.

[0185] In combination with the abnormal fluctuation identification accuracy and false positive rate, the parameter sensitivity analysis is used to dynamically adjust the discrimination threshold of the key state turning point.

[0186] In specific implementation, the embodiment defines the abnormal fluctuation identification accuracy (Precision) and false positive rate (False Positive Rate, FPR) through the effect evaluation of the abnormal fluctuation position identification after the historical medicament is put in:

[0187]

[0188] Among them, TP, TP, and TN represent the number of real abnormal positions, the number of false abnormal positions, and the number of correctly identified normal state positions, respectively.

[0189] Further, the embodiment performs parameter sensitivity analysis on the discrimination threshold (i.e., the sustained stability threshold R th and the boundary clear degree threshold B th ) of the key state turning point in step S102.

[0190] (1) Different discrimination threshold combinations are used to re-analyze the data after the action of the historical medicament, the corresponding Precision and FPR are calculated, and the relationship surface between the threshold and the identification effect is established.

[0191] (2) The optimization method (such as genetic algorithm) is used to select the discrimination threshold combination that can maximize the Precision and minimize the FPR, and the optimized R th,opt , B th,opt is obtained.

[0192] The above process realizes the dynamic adjustment of the selection conditions of the key state turning point to adapt to the state space variation characteristics after the action of the medicament.

[0193] According to the dynamically adjusted discrimination threshold, the key state turning point is re-identified, and the adaptive optimization of the state triggering threshold and the precise medicament putting control strategy is realized.

[0194] In specific implementation, the determined optimized discrimination threshold combination is used to re-identify the key state turning point of the data after the action of the medicament.

[0195] Further, the process of steps S103 and S104 is implemented again according to the re-identified key state turning point to re-determine the optimized dominant state transition path and state trigger threshold interval, and a complete closed-loop adjustment is completed.

[0196] Finally, through the above dynamic adaptive process, the embodiment can realize continuous optimization of the precise dosing strategy, so that the timing, position and dosage of dosing always match the actual state parameter change trend, thereby ensuring long-term stable and efficient operation of the sewage treatment process.

[0197] Embodiment 2

[0198] As shown in Figure 2 The embodiment disclosed herein provides a sewage biochemical treatment dosing control system based on artificial intelligence, which comprises:

[0199] The trajectory construction module 201 is configured to acquire state parameter data continuously collected at multiple points in the treatment area, construct a state space according to the correlation between the parameters, calculate the fractal dimension of the state space, and form a fractal evolution trajectory of the state space.

[0200] The abnormal point identification module 202 is configured to identify multiple abnormal fluctuation positions based on the morphological features of the fractal evolution trajectory, and determine the key state turning point according to the spatial density of the abnormal fluctuation positions.

[0201] The network construction and path searching module 203 is configured to construct a feature network with the key state turning points as nodes according to the distribution mode of the key state turning points, and identify a dominant state transition path from the feature network.

[0202] The threshold control and dosing module 204 is configured to determine a corresponding state trigger threshold condition based on the dominant state transition path, and monitor the state space data in real time. When it is detected that the state parameter data reaches the state trigger threshold condition, the precise dosing of the corresponding position of the medicament is triggered.

[0203] The self-adjusting closed-loop module 205 is configured to re-calculate the change of the fractal dimension of the state space in the treatment area after dosing, and dynamically adjust the selection condition of the key state turning point and the state trigger threshold, so as to realize dynamic adaptive control of the precise dosing of the medicament.

[0204] The above formulas are dimensionless numerical calculations, and the formulas are obtained by software simulation of a large amount of data to obtain a formula of the nearest real situation. The preset parameters, weights and threshold values in the formula are set by a person skilled in the art according to the actual situation.

[0205] The above-described embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented by software, the above-described embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable apparatus. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center through a wired network or a wireless network. The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0206] The above merely describes certain exemplary embodiments of the present application by way of illustration, and it is needless to say that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present application. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of the claims of the present application.

Claims

1. A wastewater biochemical treatment dosing control method based on artificial intelligence, characterized in that, include: S101: Acquire state parameter data collected continuously from multiple points within the area to be processed, construct a state space based on the correlation between the parameters, calculate the fractal dimension of the state space, and form the fractal evolution trajectory of the state space. S102: Identify multiple abnormal fluctuation locations based on the morphological characteristics of the fractal evolution trajectory, and determine key state inflection points by the spatial density of the abnormal fluctuation locations; S103: Based on the distribution pattern of key state inflection points, construct a feature network with key state inflection points as nodes, and identify the dominant state transition path from the feature network. S104: Based on the dominant state transition path, determine the corresponding state trigger threshold condition, and monitor the changes in state space data in real time. When the state parameter data reaches the state trigger threshold condition, trigger the precise delivery of the agent at the corresponding location. S105: After the agent is administered, the fractal dimension change of the state space in the area to be treated is recalculated, and the selection conditions of key state inflection points and state trigger thresholds are dynamically adjusted to achieve dynamic adaptive control of precise agent administration.

2. The wastewater biochemical treatment dosing control method based on artificial intelligence according to claim 1, characterized in that, The construction of the state space, calculation of its fractal dimension, and formation of its fractal evolution trajectory include: The spatiotemporal coupling window length of the state parameters is determined based on the acquisition frequency and spatial distribution characteristics of the parameters within the region to be processed. Within the spatiotemporal coupling window, the temporal fluctuation characteristics and spatial distribution difference characteristics of the state parameters are extracted to form a multi-scale feature set; The non-stationary correlation index between state parameters is calculated based on multi-scale feature sets to determine the dynamic correlation structure of state parameters. A state space is constructed based on the dynamic correlation structure, and the fractal dimension is calculated based on the irregularity of the parameter trajectory in the state space to obtain the fractal evolution trajectory of the state space.

3. The wastewater biochemical treatment dosing control method based on artificial intelligence according to claim 2, characterized in that, The dynamic association structure for determining state parameters includes: The time-varying frequency domain coherence between different state parameters is extracted using the sliding window cross-spectral analysis method. Time-frequency coherent spectra are generated based on time-varying frequency domain coherence, and highly coherent regions in the spectra are extracted. The significance weights of parameter associations in the dynamic association structure are determined based on the duration and intensity of the high coherence region. The selection of state-space parameters is optimized based on significance weights, and the correlation of state parameters is characterized by dynamic correlation structure.

4. The wastewater biochemical treatment dosing control method based on artificial intelligence according to claim 3, characterized in that, The determination of key state inflection points includes: Based on the duration and magnitude of local mutations in fractal evolution trajectories, the locations of abnormal fluctuations can be identified. For the local abrupt change characteristics of the abnormal fluctuation location, calculate the spatial local density and spatial distribution inhomogeneity index; Two-dimensional clustering is performed using spatial local density and spatial distribution non-uniformity index to obtain high-density areas at locations of abnormal fluctuations; Key state inflection points are determined based on the sustained stability of high-density areas and the clarity of boundaries between regions.

5. The wastewater biochemical treatment dosing control method based on artificial intelligence according to claim 4, characterized in that, The construction of a feature network with key state transition points as nodes, and the identification of dominant state transition paths from the feature network, includes: Calculate the propagation probability and time delay characteristics of critical state inflection points in the state space trajectory; An initial association network between inflection points is constructed based on the characteristics of transmission probability and time delay, and the importance of nodes is characterized by node degree and clustering coefficient. The initial association network is weighted based on node importance index, and the feature network structure is optimized using link prediction method; Based on the optimized feature network structure, the stability and uniqueness of the inter-node association paths are calculated, and the dominant state transition paths are identified.

6. The wastewater biochemical treatment dosing control method based on artificial intelligence according to claim 5, characterized in that, The optimization of the feature network structure using the link prediction method includes: Construct an initial link feature matrix based on the historical transmission records between state transition points; The probability distribution of potential connections between nodes is obtained by using a random walk algorithm, and the predicted connection strength is calculated by combining the link feature matrix. Based on the predicted connection strength, the links in the initial network of connections between nodes are enhanced or suppressed to optimize the network structure; The optimization effect is evaluated based on the difference in path connectivity before and after network structure optimization, and the final feature network structure is determined.

7. The wastewater biochemical treatment dosing control method based on artificial intelligence according to claim 6, characterized in that, The evaluation of optimization effects includes: Based on the dominant state transition paths before and after feature network structure optimization, calculate the degree of path coverage of abnormal fluctuation locations. The changes in the accuracy of the feature network in identifying key state inflection points were determined after the coverage was increased; Based on changes in identification accuracy, the correlation between the precision of drug delivery and the optimization and adjustment of key state inflection points is quantified. The overall improvement effect of the feature network optimization adjustment on the state transition monitoring effect is evaluated based on the correlation assessment, and the optimization effect is completed.

8. The wastewater biochemical treatment dosing control method based on artificial intelligence according to claim 7, characterized in that, The determination of the corresponding state trigger threshold condition includes: Based on the distribution characteristics of state parameters of nodes in the dominant state transition path, the extreme values ​​and rates of change of state parameters are statistically analyzed. Based on the extreme values ​​and rates of change of the state parameters, an initial state trigger threshold range is constructed, and the warning sensitivity of each threshold range is determined. Based on the degree of consistency between the early warning sensitivity and the historical events of actual state changes, the threshold range is dynamically adjusted using a multi-objective optimization method. When the rate of change of state parameters within the dynamically adjusted threshold range meets the preset triggering conditions, the state triggering threshold condition is determined and the drug delivery is triggered.

9. The wastewater biochemical treatment dosing control method based on artificial intelligence according to claim 8, characterized in that, The selection criteria for dynamically adjusting key state inflection points include: Based on the state-space fractal dimension change curve after drug administration, the time-delay characteristics and spatial diffusion characteristics of drug action are identified. Based on the time delay and spatial diffusion characteristics of the drug's action, the spatial density criterion in abnormal fluctuation identification is adjusted. By combining the accuracy and false alarm rate of abnormal fluctuation identification, the discrimination threshold of key state inflection points is dynamically adjusted using parameter sensitivity analysis. By re-identifying key state inflection points based on dynamically adjusted discrimination thresholds, adaptive optimization of state trigger thresholds and drug delivery precision control strategies can be achieved.

10. A wastewater biochemical treatment dosing control system based on artificial intelligence, implemented based on the wastewater biochemical treatment dosing control method based on artificial intelligence as described in any one of claims 1-9, characterized in that, include: A trajectory construction module is used to acquire state parameter data continuously collected from multiple points within the area to be processed, construct a state space based on the correlation between the parameters, calculate the fractal dimension of the state space, and form the fractal evolution trajectory of the state space. The anomaly detection module is used to identify multiple abnormal fluctuation locations based on the morphological features of the fractal evolution trajectory, and to determine key state inflection points by the spatial density of the abnormal fluctuation locations. The network construction and path finding module is used to construct a feature network with key state inflection points as nodes based on the distribution pattern of key state inflection points, and to identify the dominant state transition path from the feature network. The threshold control delivery module is used to determine the corresponding state trigger threshold conditions based on the dominant state transition path and monitor the changes in state space data in real time. When the state parameter data is detected to reach the state trigger threshold conditions, the precise delivery of the agent at the corresponding location is triggered. The self-adjusting closed-loop module is used to recalculate the fractal dimension change of the state space in the area to be processed after the drug is administered, and dynamically adjust the selection conditions of key state inflection points and state trigger thresholds to achieve dynamic adaptive control for precise drug administration.

Images